What a Good Instrument Cannot Fix

The design was admired well beyond the field, and its influence is easy to trace: the Princeton programme a decade later, and the whole subsequent literature of micro-psychokinesis, are variations on Schmidt's machine. What the apparatus could not do was constrain the experimenter's decisions about the experiment. Nothing in a decay-driven generator determines how many trials to collect before stopping, which subjects to keep running, whether to report a series that went nowhere, or which of several possible analyses to present. Those choices sit outside the hardware, and they are where the field's difficulties turned out to live. The point was made decisively by the 2006 meta-analysis of the whole random generator literature, which found that the reported effect size declined as studies grew larger, with the biggest deviations coming from the smallest experiments.

If the effect were physical, size of study should not predict its magnitude. If the literature is being filtered by what gets written up, that is precisely the pattern expected. Schmidt continued to publish and never accepted this reading, and the argument over the correction applied in that analysis is genuine and technical. The asymmetry itself is not disputed by anyone.